Analysis of Methods for Evaluating Responses of LLMs in Retrieval-Augmented Generation
Abstract
Retrieval-Augmented Generation (RAG) systems are increasingly employed to provide domain-specific, context-aware answers by combining large language models (LLMs) with external knowledge sources. However, evaluating the quality of chatbot responses, particularly when underlying document sets evolve over time, remains a significant challenge. In this paper, we investigate methods for assessing changes in RAG-generated answers before and after an update of the underlying knowledge base. We construct a dataset of paired responses (baseline vs. updated) and explore a spectrum of evaluation strategies: from modern LLM-based judgments to lightweight small language models (SLMs) and classical pre-trained architectures such as BERT. Our study focuses on the central question: Is the updated response better, worse, or equivalent compared to the baseline? We report experimental results across different models, highlighting strengths, weaknesses, and biases in automated comparative evaluation. The findings provide insights into practical approaches for response quality assessment in evolving RAG systems, offering a foundation for more robust and scalable evaluation pipelines in real-world applications.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.